DE102018219290B4 - Method for teaching a personalized headlight device of a motor vehicle - Google Patents

Abstract

Method for teaching a personalized headlight device (18) of a motor vehicle (18), wherein several different light distributions can be set using the headlight device (18), by carrying out the following method steps by a learning unit (LE) which has an artificial neural network (NN):a ) Learning a user selection of a respective light distribution on the headlight device depending on at least one traffic situation (S1), b) Determining a current traffic situation (S2), c) Setting one of the light distributions on the headlight device depending on the learned user selection and from the current traffic situation (S3), - setting a new light distribution on the headlight device (18) by the learning unit (LE) for learning and / or inducing a further user selection, and / or - checking after setting one of the light distributions (S3) within a predetermined time interval and/or after a change in the current traffic situation by the learning unit (LE), whether the user selects a light distribution other than that set by the learning unit (LE) within the predetermined time interval, and in the case of a different selected light distribution, the learning unit (LE) carries out the method steps a) to c) again and - detecting every user-side change in the light distribution after setting one of the light distributions for a predetermined period of time by means of a respective correction index by the learning unit (LE) and iteratively carrying out the method steps a) to c) through the learning unit until a number or a sum of the correction indices becomes minimal.

Description

The invention relates to a method for teaching a personalized headlight device of a motor vehicle, with the headlight device being able to be used to set several different light distributions. The method is carried out by a learning unit that has an artificial neural network. The invention also relates to a lighting device with a headlight device for generating multiple light distributions.

Many modern motor vehicles have a segmented headlight device or a pixel light system that can be used to implement adaptive lighting systems. Various sensor data can be taken into account. Dozens of parameters can be incorporated into the generated control image or light distribution. In many cases, camera data is also evaluated and taken into account.

So far it is not known from the applicant side how the light distribution can be adapted to the driver's switching on, off and dimming habits. This means that the light distribution cannot be personalized for certain situations. Furthermore, there is no possibility of detecting the driver's interventions and the automatic lighting system and using this to generate a personalized automatic lighting system. Most of the time there is only a standard automatic light control for a normal user, although this standard automatic light control can differ greatly depending on individual preferences. This means that while a driver can generally customize the standard automatic headlight system, it is not possible to use it to provide a complete personalized headlight setup.

The disclosure document DE 10 2011 100 277 A1 describes a method for parameterizing an assistance system of a motor vehicle, in which at least one presented parameter for controlling at least one assistance function of the assistance system is automatically adapted to the individual driving behavior of a driver of the motor vehicle.

The disclosure document DE 10 2005 036 002 A1 relates to a method for controlling a lighting device of a motor vehicle driven by an operator. The method is characterized by the fact that the operator's line of sight towards the vehicle in front of him is recorded. Depending on the recorded information about the viewing direction, at least one parameter of an adjustable and adjustable headlight of the lighting device is set.

The disclosure document DE 10 2007 009 895 A1 describes a control device for controlling a vehicle unit. The control device has a situation detection unit, a vehicle unit operating model storage unit, an inference execution unit, a vehicle unit control unit, an operation detection unit and a model learning unit. The model learning unit may learn the vehicle unit operation model stored in the vehicle unit operation model storage unit by using a situation detected by the situation detection unit. A situation can be learned where operational needs arise. The operating requirements relate in particular to a vehicle unit.

It is the object of the present invention to provide a headlight device or an associated method which makes it possible to generate personalized light distributions.

This task is solved according to the independent patent claims of this application. Useful further developments, further embodiments and other variants of the invention result from the subclaims, the description and the figures.

The invention provides a method for teaching a personalized headlight device of a motor vehicle. Using the headlight device, several different light distributions can be adjusted. The method is carried out by a learning unit that has an artificial neural network. In step a), a user selection of a respective light distribution on the headlight device is learned depending on at least one traffic situation. The term learning can also include the meaning of learning or studying. For example, the artificial neural network can learn this selection by observing a person making the user selection. In this context, learning can also mean that a user selection is recorded, registered and/or saved. The user's selection can be recorded by evaluating corresponding input signals or by observing the person. The input signals can, for example, be designed as pressing a button or saying a keyword. For example, if a person says the phrase “adjust the city light,” the learning unit can record this phrase and learn from it that the person has set a certain light distribution would like to have. This makes it possible for the learning unit to record the user's selection and thus learn or learn the user's selection of one of the respective light distributions. At least one traffic situation is taken into account.

The at least one traffic situation can be specified and provided to the learning unit. For example, the learning unit can access data relating to at least one traffic situation. This data can be stored, for example, in a backend or an external server. It is also possible for the learning unit to access data from a sensor device that detects the surroundings of the motor vehicle. The learning unit can evaluate and analyze this data accordingly. With the help of this analysis of the data from the sensor device, the learning unit can determine the at least one traffic situation. This means that, on the one hand, the learning unit is able to learn the user's selection and, on the other hand, the learning unit can also be able to determine at least one traffic situation.

In step b), a current traffic situation is determined. The current traffic situation can be characterized in particular by the fact that it takes place at a different time than the at least one traffic situation. With regard to determining the current traffic situation, the learning unit can act or proceed in the same way as when determining the at least one traffic situation. In particular, step a) can extend over a predetermined period of time. The at least one traffic situation and/or current traffic situation can, for example, relate to a road course such as curves or intersections or the presence of traffic signs. For example, the learning unit can use a vehicle camera to record and recognize traffic signs and take these traffic signs into account when determining traffic situations.

In a further step c), one of the light distributions on the headlight device is adjusted depending on the learned user selection and the current traffic situation. This means that the learning unit can learn how to manually adjust a driver's light distributions and thereby take at least one traffic situation into account. In a current traffic situation, the learning unit can rely on this learned setting behavior of the driver and adjust the light distribution depending on the learned user selection and the current traffic situation. This allows an individual automatic lighting system to be created. This makes it possible to provide personalized lighting that supports the driver's habits and thus increases safety. Furthermore, the driver's sense of comfort can be increased because he no longer has to adjust the light distributions manually. This means he can concentrate more on maneuvering the vehicle. The user selection can, for example, be done manually or voice-controlled. With the help of this method it is possible to at least reduce the number of user selections.

A further variant of this invention provides that, in order to learn the user's selection, a person is observed visually and/or acoustically while maneuvering the motor vehicle. In many cases, the person is also a driver of the motor vehicle. The person can be visually observed using a microphone, for example. For this purpose, for example, sound waves can be recorded in order to record and/or analyze the person's voice commands. In this way, set light distributions, which are set using voice control, can be recorded and thus learned by the learning unit. Optical observation can also be carried out in a similar way. For example, interior vehicle cameras can be used to observe or record the person or driver. In particular, it can be recorded where a driver's hand is located. With the help of this configuration, both linguistic settings and manual settings of the driver on the automatic lighting system can be observed. This can be used to learn the user's selection of the respective light distribution.

A further embodiment provides that a position of a person's hand is detected and/or an analysis of acoustic waves is carried out for a language position of the headlight device in order to register the user's selection of the respective light distribution for learning. If the position of the person's hand is recorded, the learning unit can determine whether the light distribution is being adjusted or is about to be adjusted. In particular, an impending change in the light distribution can be recognized by the learning unit. In this case, the learning unit could switch to a recording mode, for example, focusing on observing or registering the upcoming user selection. This can mean in particular that the learning unit allocates its internal resources for learning the user's selection. This can also be done using a keyword. For example, the learning unit registers a keyword that is an upcoming If an existing change in the light distribution is indicated, the learning unit can internally release or make available those digital resources of the learning unit for learning the user's selection. This can ensure that the process of learning or learning can be carried out effectively.

A further embodiment of this invention provides that the at least one traffic situation and/or the current traffic situation is determined by a visibility distance for the person, a state of health of the person, a position and orientation of the motor vehicle and/or a road course for the motor vehicle. For example, the vision of a normal person can be used to determine the person's range of vision. If personal data regarding the visibility is known or available, the learning unit can also determine an individual visibility for the person. Different traffic situations can also be taken into account. A traffic situation can also be determined depending on the person's state of health. The person's state of health can therefore influence at least one traffic situation as well as the current traffic situation. Personal information, if available and available, can also be taken into account.

A position and orientation of the motor vehicle can be determined in particular using a sensor device of the motor vehicle. Such information can be transmitted to the learning unit, for example, using a CAN bus network of an on-board electrical system of the motor vehicle. The course of the road can take on several different forms. This can address curves, intersections, pedestrian crossings and traffic signs. For example, an analysis of traffic signs can also be used to identify the future course of the road.

This can also be achieved using predictive route data. Predictive route data can in particular be provided on a backend server and the learning unit can thus take a future road course into account when determining the traffic situation. The same applies analogously and analogously to the case in which the learning unit analyzes recorded traffic signs. This makes it possible to record or describe different traffic situations in detail. This can help provide a personalized headlight setup. Since detailed traffic situations can be determined in this way and at the same time the user's selection or manual adjustment behavior is determined, the learning unit can develop and provide personalized light distributions tailored to the driver. With the help of the neural network, the learning unit can learn an optimal or personalized light distribution for every traffic situation. For example, if the driver fades the light up or down several times in a certain traffic situation or at a certain point, the learning unit can recognize this and save the driver from having to intervene in the future.

A further embodiment provides that after setting one of the light distributions within a predetermined time interval and/or after a change in the current traffic situation, the learning unit checks whether the user has selected a light distribution other than the one set by the learning unit within the predetermined time interval, and in In the case of a different selected light distribution, the learning unit carries out process steps a) to c) again. The predetermined time interval can, for example, include a maximum period of 15 seconds, in particular 10 seconds, preferably 5 seconds. This embodiment of the invention can therefore provide that the learning process for the learning unit or the method is not completed after a first run. For example, if the learning unit has set a new light distribution, the learning unit can check whether the set light distribution is changed by the user within the specified time interval. If a change occurs within the specified time interval, this could indicate that the learning unit has not yet been optimally trained.

However, it is also possible that the person's or driver's habits have changed. For this reason, this embodiment provides in particular that if the user changes the set light distribution within the specified time interval, the learning unit carries out the method steps again. This embodiment can therefore be viewed as a further method step in which the learning unit checks or monitors whether the set light distribution matches the learned user selection of the light distributions. To do this, the learning unit can determine corresponding values relating to the user's selection and compare these values with one another. These values can refer to a further specified period of time. The values can represent an adjustment habit of the driver and can therefore be referred to as a “habitual value for the headlight device”.

Ideally, no changes are made by the user after the light distribution has been adjusted by the learning unit. This can be done through the learning process can be seen as confirmation that it has learned the user behavior correctly. This check can be used to ensure that the learning unit can adapt flexibly to the driver's changing habits.

A further embodiment provides that the learning unit detects every change in the light distribution by the user for a predetermined period of time after setting the light distributions using a respective correction index and the learning unit is designed to carry out the method steps a) to c) iteratively until a number or a sum of the correction indices becomes minimal, in particular 0. This embodiment therefore provides in particular that a respective correction index is recorded or determined after every change in the light distribution by the user. This applies for the specified period of time. The specified period of time can, for example, cover one or more months. The correction index can also be designed as a weighted correction index. In addition to the presence of a user-induced change in the light distribution, the weighted correction index can also include the extent or the extent of the change in the light distribution. This makes it possible to determine whether the user change represents only a minor modification or whether the change to the set light distribution is of a greater magnitude. Ideally, the learning unit sets the light distributions so that no user changes are necessary. In this case, no user changes are recorded and therefore no correction index is recorded. It can be provided that an initial correction index of 0 is provided. In an ideal learning unit that is optimally tailored to the driver, the correction index would be 0.

In this embodiment, the learning unit can iteratively improve until the number or sum of correction indices becomes minimal. For example, it can be determined how well the learning unit has adapted to the driver's manual adjustment behavior with regard to the light distribution. In this case, it can be determined how well the learning unit has adapted to the driver. The number or the sum of the correction indices can be used as a quality measure. In this configuration, a lower correction index tends to mean a more personalized learning unit. In the case of weighted correction indices, the extent of user-made changes to the set light distribution can be taken into account. In this case, a slightly shifted light distribution would have a lower correction index than if the driver changed a high beam to a low beam. With the help of the correction index, a comprehensible quality criterion can be created, which makes it possible to determine the flexible adaptability of the learning unit to the person. This can help to optimize the neural network or the learning unit with regard to learning the user's selection.

The learning unit can use an artificial neural network that is designed to use reinforcement learning to reduce the number of correction indices. This can mean in particular that the correction indices represent information that provides informative feedback for the learning unit. This means that the learning unit can use the correction indices to receive feedback indicating whether the light distribution has been set correctly. The learning unit can therefore use the correction indices as informative feedback to validate the set light distribution. For this purpose, the learning unit can in particular use specially developed neural networks. For example, a long short-term memory network can be used. Such a neural network can be trained using unsupervised or supervised data. Long-short-term memory networks or LSTM cells can access not only information from a point in time in the past, but also information from points in time further in the past. Thus, it is possible that LSTM cells can take information into account at all previous time points. If the times are assigned to different layers of the neural network, LSTM cells cannot access the layer immediately before, but also information from layers further back. In particular, this makes it possible to record a chronological sequence of changes made by the user as well as the associated traffic situations.

This data can be collected in the motor vehicle and fed to the learning unit for learning the user's selection. It can also be provided that this data is stored and processed in a backend. This is particularly advantageous if analyzing this data is too computationally intensive for the neural network of the learning unit. In particular, it is possible for the learning unit to create a driver profile that combines the user's selection of the driver with the corresponding traffic situations. These examples show how the learning unit's learning of user selection can be made even more effective to provide the personalized headlight device.

A further embodiment provides that an identity of the person is recorded and the identity is linked to the user selection. The person's identity can be determined using image data, for example. It can also be provided that the person is recognized using voice analysis, a cardiogram, a fingerprint or other biometric data. The biometric data can include, for example, a gait profile, an individual behavior of the person and/or a fingerprint. This allows the learning unit to learn the user's selection of the person and at the same time link the user's selection to the person's identity. The learning unit can thus generate or create an individual driver profile by storing the person's user behavior. This driver profile can be assigned to the person based on the recorded identity. This means, for example, that a driver profile can be transferred to another motor vehicle. In this case, a further learning unit would not have to learn the person's individual attitude behavior to the same extent as the original learning unit did.

A further embodiment provides that the at least one traffic situation is divided into different classes and the current traffic situation is assigned to one of these classes, and the light distribution is adjusted depending on this class. Each class can in particular represent traffic situations that have a predetermined degree of similarity to one another. For example, driving straight ahead can be classified into the same class at different locations. The learning unit can thus analyze the different traffic situations to see to what extent the differences in these traffic situations represent a deviation that makes it necessary to provide a new class for traffic situations.

For example, straight-ahead driving during a day and straight-ahead driving during a night would be assigned to different classes. In this case, the learning unit could recognize that the different points in time certainly mean a change that is relevant with regard to an assessment of the traffic situation. Due to the different times, the visibility for the driver, for example, is different. This can result in straight-ahead driving during the day being classified in a different class than straight-ahead driving during the night. This means that the learning unit can analyze complex traffic situations more precisely and the learning unit can recognize differences in the traffic situations. In addition, the learning unit can determine the relevance of these differences. For this purpose, the learning unit can in particular determine a similarity standard that states to what extent the different traffic situations are comparable. In this way, the learning unit can combine similar traffic situations in one class.

During operation of the motor vehicle, other traffic situations may occur, which the learning unit can also take into account. These new traffic situations can either be assigned to an already existing class or the new traffic situation can require the creation of a new class. For example, if a driver drives into a traffic-calmed zone for the first time in which driving is only permitted at walking speed, this represents a new traffic situation. If the learning unit registers this new traffic situation for the first time and the learning unit cannot assign this new traffic situation to existing classes , the learning unit can provide a new class and assign the new traffic situation to this new class. The light distribution can then be adjusted depending on the classes or the applicable class. This can make determining the current traffic situation much easier or faster. This eliminates the need for a complex analysis of traffic situations with regard to unnecessary details.

For example, one class can be intended for straight-ahead driving and another class for cornering. Further classes can distinguish whether the motor vehicle is maneuvered within a town or outside a town. In particular, a well-trained neural network can divide different traffic situations into more classes than an untrained neural network. For example, four different traffic classes can be provided. The first class could mean a straight drive, the second class could mean a curve, the third class could mean a trip within a town and the fourth class could mean a ride outside the town. A combination of these classes is also possible.

For example, cornering outside the town could be assigned to second and fourth class at the same time. This means that another class can be formed from the original basic classes. Additional classes can be formed, for example, by a combination of original basic classes. For example, a fifth class could be formed from a combination of the second and fourth classes. In this case, this would correspond to cornering outside the town. The different classes can be predefined or learned by the neural network itself. This means that the neural network can take new traffic situations into account by introducing new classes. By dividing traffic situations into appropriate ones Classes, the computational effort for the neural network can be significantly minimized. This means that not every minimal deviation necessarily leads to a new traffic situation, but similar traffic situations can be treated in the same way. To do this, the neural network can compare two traffic situations using a corresponding correlation as a measure of similarity between the traffic situations. The neural network or learning unit can quantitatively evaluate how similar two traffic situations are to each other and, based on this, assign these two traffic situations to a single class or two different classes.

A further embodiment provides that the learning unit sets a new light distribution on the headlight device for learning and/or inducing a further user selection, with a previous user selection being ignored, in particular for this setting of the new light distribution. This means that the learning unit can be designed to make targeted changes to the light distribution. The learning unit can thus set a new light distribution on the headlight device in an upstream step of the method. Previous user selections can be completely ignored. This upstream procedural step makes sense if the learning unit would like to test whether the driver would prefer a different light distribution than previously usual, or if the learning unit would like to provoke a user selection by setting a new light distribution in advance. This means that the light distribution can also learn whether the driver would accept light distributions other than those previously selected or set. If a change is made by the user after the new light distribution has been specifically set by the learning unit, this can be considered a user selection. This ensures that the learning unit does not automatically develop an automatic lighting system that does not suit the driver. At the same time, the driver can be offered new light distribution, which he can accept or reject. If the driver rejects the light distribution presented to him by the learning unit, he can do this by appropriately changing the light distribution preset by the learning unit.

The learning unit can deliberately forget a previous user selection of the person. This is sometimes referred to as “motivated forgetting” or “consiously forgetting.” This type of forgetting differs significantly from unwanted forgetting, also known as “catastrophic forgetting”. The learning unit is preferably designed so that the user's selection of the person is only consciously and not unintentionally forgotten. Forgetting can mean deleting a saved user selection of the person. The learning unit can therefore delete or no longer take into account that part of its assigned storage unit that has the user-time selection that should be forgotten. For this purpose, the learning unit or the neural network can have algorithms that make it possible to objectively determine when a user-made selection should be forgotten or deleted. For example, in this case, the learning unit can determine a discrepancy that exceeds a corresponding threshold value based on a comparison of several user selections at different time intervals. This discrepancy can be determined, for example, in the form of the similarity measure or a value based on a correlation. The learning unit can decide which time interval is affected by forgetting or deleting the associated user selections. In this way, it can be avoided that the person's previous behavior does not make learning, learning or learning the learning unit unnecessarily difficult.

A further embodiment provides that the new light distribution is evaluated as a user selection if the user does not make any changes to the new light distribution. In this case, the learning unit suggests a light distribution that is accepted by the driver. This means that the user does not make any changes to the new light distribution. This embodiment now provides that this passive behavior of the driver, i.e. the absence of a change by the user, can be viewed as consent for the new light distribution. In this case, the new light distribution is considered a user selection. It can be treated in the same way as if the new light distribution had been set by the driver. For example, the learning unit can introduce the driver to a new lighting concept that was previously unknown to the driver and can be accepted by the driver. This can help ensure that the automatic lighting system of a headlight device can be personalized even better.

A further embodiment of this invention provides a lighting device with a headlight device for generating multiple light distributions. This headlight device can be designed as a segmented headlight or pixel light system. In particular, the headlight device can have a large number of MEMS or DMD mirrors. This allows different light distributions to be set pixel by pixel. The number of pixels to be set can be several thousand pixels. The pixel light system can be implemented using an LED matrix. This LED matrix can be area modulated be bar, which means that individual areas of the light distribution can be modulated.

The pixel light system can also be designed as an LCD pixel light system. A laser pixel light system is also possible as a headlight device.

The lighting device has an interface for receiving data relating to at least one traffic situation and a current traffic situation. This interface serves in particular to transmit this data to the learning unit or to enable the learning unit to access this data. Furthermore, the lighting device has a learning unit for learning a user selection of a respective one of the light distributions on the headlight device depending on the at least one traffic situation and for adjusting one of the light distributions on the headlight device depending on the learned user selection and on the current traffic situation. Adjusting one of the light distributions on the headlight device can mean that the learning unit generates control signals which are designed to change the light distribution that is generated by the headlight device. The learning unit can therefore also be viewed as a control unit for the headlight device. The examples and advantages mentioned in the previous refinements and variants apply mutatis mutandis and analogously to the lighting device. In particular, the lighting device has physical features that make it possible to carry out the method steps. This means in particular that the learning unit can have corresponding digital resources such as a microprocessor, a storage medium, access to an on-board network of the motor vehicle and so on.

A further embodiment provides that the lighting device has a detection unit and the learning unit has an artificial neural network and the learning unit is designed to determine the traffic situations and/or the current traffic situation. The detection unit can be designed as a sensor device of a motor vehicle. Thus, for example, a camera or a microphone can be part of the detection unit. The data relating to the at least one traffic situation or the current traffic situation can be raw data, for example. These can be, for example, ultrasound signals or image data. The learning unit can analyze this information and determine a traffic situation as a result. When analyzing image data, for example, different image processing algorithms can be used. In particular, the artificial neural network can independently carry out the analysis of data relating to traffic situations. In the simplest case, this data has already been prepared, for example by being analyzed in advance on a backend server. In this case, the learning unit can access or receive this data via the interface. It is therefore possible for the learning unit to retrieve or read out the data relating to the at least one traffic situation or the current traffic situation or, in the case of raw data, to analyze this raw data accordingly and independently determine the traffic situation.

A further embodiment provides a lighting device with a detection unit for learning the user's selection, the detection unit in particular having a vehicle interior camera and/or a microphone. In this case, the detection unit has sensors which can monitor or observe, in particular, an interior of the motor vehicle. The person or their hand can be observed using the vehicle's interior camera. The microphone can be used to analyze the person's spoken text. In particular, voice commands can be received and evaluated by the learning unit. The learning unit can thus learn the user's selection of the driver using the two components of the detection unit. In this embodiment of the invention, the advantages and examples already mentioned also apply mutatis mutandis to this embodiment.

A further embodiment provides a motor vehicle with a lighting device. The learning unit can therefore always provide a personalized light distribution tailored to the driver while the motor vehicle is driving. Here too, the advantages and examples already mentioned apply analogously to the motor vehicle.

A further embodiment of this invention provides a computer program product which includes instructions that cause the lighting device to carry out method steps a) to c).

This invention also provides a computer-readable medium on which the computer program product is stored.

A method described in this application may also be in the form of a computer program product that implements the method of the learning unit when executed on the learning unit. Likewise, an electronically readable data carrier with data stored on it can be used electronically readable control information is present, which includes at least one described computer program product and is designed in such a way that it carries out a described method when using the data carrier in a learning unit. The invention also includes the learning unit for the motor vehicle. The learning unit can have a processor device that is set up to carry out an embodiment of the method. For this purpose, the processor device can have at least one microprocessor and/or at least one microcontroller. Furthermore, the processor device can have a program code that is designed to carry out the embodiment of the method when executed by the processor device. The program code can be stored in a data memory of the processor device.

The invention will now be explained in more detail with reference to the accompanying drawings. Possible or preferred embodiments of the invention are described in the figures. The examples shown in the figures are exemplary and therefore not restrictive.

The invention also includes developments of the method according to the invention, which have features as have already been described in connection with the developments of the motor vehicle according to the invention. For this reason, the corresponding developments of the method according to the invention are not described again here.

The invention also includes the combinations of the features of the described embodiments.

• 1 eine schematische Seitenansicht auf ein Kraftfahrzeug mit einem darin sitzenden Fahrer; und
• 2 ein schematisches Ablaufdiagramm eines beispielhaften Verfahrens.

Examples of embodiments of the invention are described below. This shows:

• 1 a schematic side view of a motor vehicle with a driver sitting in it; and
• 2 a schematic flowchart of an exemplary method.

The exemplary embodiments explained below are preferred embodiments of the invention. In the exemplary embodiments, the described components of the embodiments each represent individual features of the invention that are to be considered independently of one another and which also further develop the invention independently of one another. Therefore, the disclosure is intended to include combinations of the features of the embodiments other than those shown. Furthermore, the described embodiments can also be supplemented by further features of the invention that have already been described.

In the figures, the same reference numerals designate functionally identical elements.

1 shows an example of a motor vehicle 10 in which a person 12 is sitting. Person 12 is in the example of 1 at the same time a driver of the motor vehicle 10. The motor vehicle 10 can in particular be designed as a motor vehicle. The motor vehicle 10 includes a lighting device 26, the lighting device 26 in turn having different components.

A headlight device 18, which is part of the lighting device 26, is arranged on a front of the motor vehicle 10. The lighting device 26 also includes a learning unit LE, a microphone 14, an interface 20, and a camera 16. In the example of 1 also via an external vehicle camera 22. The learning unit LE has a neural network NN. The learning unit LE can receive information or data from the vehicle interior camera 16, the microphone 14, the headlight device 18 and from the interface 20. At the same time, the learning unit LE can generate control signals and use these control signals to control the headlight device 18. Likewise, the learning unit LE can be designed to control the components of the detection unit, namely the vehicle interior camera 16 and the microphone 14. This may be necessary, for example, if the vehicle interior camera 16 needs to be realigned due to a changed seating position of the person 12. The same applies to the microphone 14. This means that the learning unit LE can also transmit control signals to the vehicle interior camera 16 and the microphone 14. The same can apply analogously to the vehicle exterior camera 22.

1 also shows a backend server 24. This backend server 24 can, for example, provide predictive route data. Further information such as a GPS signal, current information on upcoming traffic situations and other relevant data relating to at least one traffic situation or the current traffic situation can also be transmitted.

With the help of the vehicle interior camera 16, the person 12 can be observed or observed. In particular, it can be observed which changes the person 12 makes to the headlight device 18 using his hand. If the headlight device 18 also allows a change via voice commands, the learning can unit LE using the microphone 14 to record these acoustically induced changes. This enables the learning unit LE to comprehensively record the user's changes to the headlight device 18 and thus to learn the user's selection of the light distribution.

At the same time, the learning unit LE can determine at least one traffic situation using the external camera 22 and other sensors (not shown). For this purpose, the learning unit LE can access raw sensor data from these sensors. The LE learning unit can analyze this raw data accordingly and thus determine the respective traffic situation. For example, if an image from the vehicle exterior camera 22 shows a road with several wide white lines that are arranged next to each other, the learning unit LE can, for example, infer a zebra crossing. If the learning unit LE recognizes a corresponding traffic sign that indicates the zebra crossing using the vehicle exterior camera 22, the learning unit can also recognize the zebra crossing by analyzing the traffic sign. This is just a possible example of how the LE learning unit can recognize or determine a traffic situation.

For example, if the learning unit LE does not have enough digital resources to determine the respective traffic situation, this data can be transmitted to the backend server 24. There, this data relating to the respective traffic situation can be evaluated and a result of this evaluation can be communicated or transmitted to the learning unit LE via the interface 20. The learning unit LE can in particular access further sensors on a CAN bus network (not shown) of the motor vehicle 10 and use their information to determine the respective traffic situation.

2 shows an example of a possible flow chart of a process. At the beginning of a journey or when the motor vehicle 10 and the lighting device 26 are activated, an initial light distribution is generated in a step S0. This initial light distribution may not yet be sufficiently personalized in an untrained neural network NN. In a first step S1, manual interventions by the person 12 on the light distribution can be recorded. The recording of these manual interventions by the person 12 can be done using various sensors of the recording unit. Thus, in step S1, the learning unit LE can learn or learn the user's selection of a respective light distribution on the headlight device 18.

In a second step S2, the current traffic situation is determined. This can in particular be time-delayed compared to the at least one traffic situation. In step S1, the at least one traffic situation is determined; in the second step S2, the current traffic situation is determined. The learning unit LE can accomplish this by appropriately retrieving or analyzing or evaluating the data relating to the respective traffic situation.

In a further third step S3, one of the light distributions on the headlight device 18 can be adjusted depending on the learned user selection and the current traffic situation. A personalized light distribution can thus be generated in the third step S3. This means that the headlight device 18 or the lighting device 26 can be personalized. Ideally, these new set light distributions correspond to the needs of the person 12 and make further user changes by the person 12 unnecessary. The person 12 is usually the driver of the motor vehicle 10, but other people can also be used to carry out this procedure.

After the third step S3 this can be done 2 The methods shown branch off in two different directions. One direction leads back to the beginning of the procedure. This can have different causes. One possibility is that the person 12 does not accept the light distribution set by the learning unit LE and manually changes the light distribution again. Another possibility would be for the learning unit to specifically set a new light distribution on its own, even though the person 12 did not change the newly set light distribution on the user side. This targeted change in light distribution can be viewed as part of a “reinforcement learning strategy”. In this case, the process begins again, whereby this targeted light distribution changed by the light unit can be understood as an upstream process step. In this case, steps S1 to S3 are repeated. If the person 12 does not make any changes to the light distribution specifically changed by the learning unit LE, the learning unit LE can evaluate and interpret this as passive consent by the person 12. In this case, a lack of reaction by the person 12 would be interpreted as consent and the light distribution preset by the learning unit LE would be interpreted by the learning unit LE as a user selection.

However, if the user makes changes to the light distributions set by the LE learning unit, this is a user responsibility intervention in the automatic lighting system of the lighting device 26. This manual intervention or this user selection can be recorded accordingly by the learning unit LE and thus the learning unit LE can further refine its information or database regarding the user behavior with regard to setting the light distributions. The learning unit LE can use the neural network NN to consciously decide when such an incentive should be specifically set to induce a user selection. This incentive is that the learning unit LE sets a new light distribution, even though the user does not adjust the light distribution. Setting this incentive does not represent a deviation from the previous process steps, since this targeted adjustment of the light distribution can be understood as an upstream, separate process step. In the event that the person 12 changes the light distribution manually again after setting the light distribution by the learning unit, it is in particular provided that the method steps a) to c) or the first step S1 to the third step S3 are run through again.

This in 2 However, the method shown as an example can also develop towards a fifth step S5. The fifth step S5 does not represent a separate procedural step. This is the situation that the learning unit LE aims for. The method reaches the fifth step S5 when it sets a light distribution and the user does not make any changes to the light distribution within a predetermined period of time. This period can be 2 to 15 seconds. If person 12 does not change the light distribution within this period of time, the learning unit LE can interpret this as tacit consent from person 12. Depending on whether the method branches off after the third step S3 to the fifth step S5 or to the fourth step S4 or first step S1, the learning unit can record a correction index or update this correction index. For example, if the method branches off again after the third step S3 in the direction of the fourth step S4 or first step S1, provision could be made for the correction index to be counted up by 1. The correction index can thus indicate that a change was made by the user after the learning unit had set a light distribution. If no change is made by the user and the method continues in the direction of the fifth step S5, the correction index can remain unchanged or not increase. The correction index can thus be used as a benchmark as to the extent to which the learning unit LE can offer the person 12 a personalized light distribution. The respective changes made by the person 12 can be weighted and incorporated into the respective correction index. In particular, the learning unit LE can attempt to keep the number or sum of correction indices as small as possible using the neural network NN or using reinforcement learning methods. Thus, the learning unit LE can in particular be designed in such a way that it strives, if possible, to go directly from the third step S3 to the fifth step S5. This corresponds to a learning unit LE that has adapted well to the person 12, which in turn means that the person 12 or the driver no longer has to operate the sometimes complex headlight device 18. He can thus be relieved of the task of operating the headlight device 18. This can increase the driver's sense of comfort and road safety.

Claims (14)

Method for teaching a personalized headlight device (18) of a motor vehicle (18), wherein several different light distributions can be set using the headlight device (18), by carrying out the following method steps by a learning unit (LE) which has an artificial neural network (NN): a) learning a user selection of a respective light distribution on the headlight device depending on at least one traffic situation (S1), b) determining a current traffic situation (S2), c) adjusting one of the light distributions on the headlight device depending on the learned user selection and the current traffic situation (S3), - Setting a new light distribution on the headlight device (18) by the learning unit (LE) for learning and/or for inducing a further user selection, and/or - After setting one of the light distributions (S3) within a predetermined time interval and/or after a change in the current traffic situation by the learning unit (LE), check whether the user has selected a light distribution other than the one set by the learning unit (LE) within the predetermined time interval is, and in the case of a different selected light distribution, the learning unit (LE) carries out the method steps a) to c) again and - Detecting every user-side change in the light distribution after setting one of the light distributions for a predetermined period of time using a respective correction index by the learning unit (LE) and iteratively carrying out the method steps a) to c) by the learning unit until a number or a sum of the correction indices becomes minimal. Procedure according to Claim 1 , whereby a person (12) is observed visually and/or acoustically while maneuvering the motor vehicle (10) in order to learn the user's selection. Procedure according to Claim 2 , wherein a position of a hand of the person (12) is detected and / or an analysis of acoustic waves is carried out for voice control of the headlight device (18) in order to register the user's selection of the respective light distribution for learning. Method according to one of the preceding claims, wherein the traffic situations and / or the current traffic situation are determined by a visibility for the person (12), a state of health of the person (12), a position and orientation of the motor vehicle (10) and / or a road course for the Motor vehicle (10) is determined. Method according to one of the preceding claims, wherein the learning unit (LE) is designed to carry out method steps a) to c) iteratively until a number or a sum of the correction indices becomes zero. Procedure according to one of the Claims 2 until 5 , wherein an identity of the person (12) is recorded and the identity is linked to the user selection. Method according to one of the preceding claims, wherein several traffic situations are divided into different classes and the current traffic situation is assigned to one of these classes, and the light distribution is adjusted depending on this class. Method according to one of the preceding claims, wherein a previous user selection is not taken into account when setting the new light distribution. Method according to one of the preceding claims, wherein the new light distribution is evaluated as a user selection if no change is made to the new light distribution by the user. Lighting device (26). - a headlight device (18) for generating multiple light distributions, - an interface (20) for receiving data relating to at least one traffic situation and a current traffic situation, - a learning unit (LE) for learning a user selection of a respective one of the light distributions on the headlight device (18) depending on the at least one traffic situation and for adjusting one of the light distributions on the headlight device depending on the learned user selection and on the current traffic situation, where the learning unit (LE) is designed, - set a new light distribution on the headlight device (18) for learning and/or for inducing further user selection, and/or - after setting one of the light distributions (S3) within a predetermined time interval and/or after a change in the current traffic situation, to check whether the user has selected a light distribution other than the one set by the learning unit (LE) within the predetermined time interval, and in the case another selected light distribution, the learning unit (LE) carries out the method steps a) to c) again and - for a predetermined period of time, to record every change in the light distribution by the user after setting one of the light distributions using a respective correction index and to carry out the method steps a) to c) iteratively until a number or a sum of the correction indices becomes minimal. Lighting device (26). Claim 10 with a detection unit for learning the user's selection, the detection unit in particular having a vehicle interior camera (16) and/or a microphone (14). Motor vehicle (10) with a lighting device (26) according to one of Claims 10 until 11 . Computer program product comprising commands that cause the lighting device (26) of the Claim 10 the procedural steps Claim 1 executes. Computer-readable medium on which the computer program product is based Claim 13 is stored.

Images